Weitere ähnliche Inhalte Ähnlich wie From Machine Translation to Machine Interpretation - Jimmy Kunzmann (20) Mehr von TAUS - The Language Data Network (20) Kürzlich hochgeladen (18) From Machine Translation to Machine Interpretation - Jimmy Kunzmann1. © EML European Media Laboratory GmbH
Think beyond the limits!
Automatic Transcription of Speech
From Applied Research to the Market
Dr. Siegfried Kunzmann
Manager R&D
www.eml.org
2. 2 © EML European Media Laboratory GmbH Jimmy Kunzmann
European Media Laboratory (founded 1997)
• Offers: research-based products, solutions and services
Speech transcription: telephone, smartphone, media data
Spoken interaction: speech interfaces
• Portfolio Structure:
Quality
Moni-
toring
Bus.
Ana-
lytics
Cus-
tomer
Sat.
Speech
Analytics
Media
Transcription
Sub-
titling
Media
Ana-
lytics
Video
Re-
trieval
Technology
Markets
Applications
Speech Transcription
(Server)
Spoken
Interaction
(Device)
Assisted
Living
Home
&
Media
Control
Car
Control
Speech
Messaging
Social
Web
Voice
Search
LBS
FAQ
Voice
mail
SMS
e-Mail
App
3. 2 © EML European Media Laboratory GmbH Jimmy Kunzmann
European Media Laboratory (founded 1997)
• Offers: research-based products, solutions and services
Speech transcription: telephone, smartphone, media data
Spoken interaction: speech interfaces
• Portfolio Structure:
Quality
Moni-
toring
Bus.
Ana-
lytics
Cus-
tomer
Sat.
Speech
Analytics
Media
Transcription
Sub-
titling
Media
Ana-
lytics
Video
Re-
trieval
Technology
Markets
Applications
Speech Transcription
(Server)
Spoken
Interaction
(Device)
Assisted
Living
Home
&
Media
Control
Car
Control
Speech
Messaging
Social
Web
Voice
Search
LBS
FAQ
Voice
mail
SMS
e-Mail
App
4. 2 © EML European Media Laboratory GmbH Jimmy Kunzmann
European Media Laboratory (founded 1997)
• Offers: research-based products, solutions and services
Speech transcription: telephone, smartphone, media data
Spoken interaction: speech interfaces
• Portfolio Structure:
Quality
Moni-
toring
Bus.
Ana-
lytics
Cus-
tomer
Sat.
Speech
Analytics
Media
Transcription
Sub-
titling
Media
Ana-
lytics
Video
Re-
trieval
Technology
Markets
Applications
Speech Transcription
(Server)
Spoken
Interaction
(Device)
Assisted
Living
Home
&
Media
Control
Car
Control
Speech
Messaging
Social
Web
Voice
Search
LBS
FAQ
Voice
mail
SMS
e-Mail
App
5. 2 © EML European Media Laboratory GmbH Jimmy Kunzmann
European Media Laboratory (founded 1997)
• Offers: research-based products, solutions and services
Speech transcription: telephone, smartphone, media data
Spoken interaction: speech interfaces
• Portfolio Structure:
Quality
Moni-
toring
Bus.
Ana-
lytics
Cus-
tomer
Sat.
Speech
Analytics
Media
Transcription
Sub-
titling
Media
Ana-
lytics
Video
Re-
trieval
Technology
Markets
Applications
Speech Transcription
(Server)
Spoken
Interaction
(Device)
Assisted
Living
Home
&
Media
Control
Car
Control
Speech
Messaging
Social
Web
Voice
Search
LBS
FAQ
Voice
mail
SMS
e-Mail
App
6. 6 © EML European Media Laboratory GmbH Jimmy Kunzmann
Real-time Speech Recognition: Applications
Audio Streaming Partial Results Infirm Words
Web-Dictation
Voicemail-to-Text Reply
Dictation-Client
Voice Messaging App (local & server)
Zawatzky CenterVoice
Broadcast News Online
Comfort functions
Voice Search
House control
7. 7 © EML European Media Laboratory GmbH Jimmy Kunzmann
Summary
• Focus of the European Media Laboratory
- Provide robust transcription technology, tools, platform
- Partner to expand language portfolio
- License tools, provide expertise, collaborate
- Provide EML transcription technology to system integrators
- Enable on-premise installation und customization
• Speech communication applications are working
- Access to lots of REAL application data helps
- Rapid customization to application domains recommended
Thank You ☺
8. 8 © EML European Media Laboratory GmbH Jimmy Kunzmann
Eureka/ZIM/OCS Project: Mediatranslator
• Mediatranslator: real-time video translation system for the financial market
• Project duration: 2015-2017
• : http://www.eurekanetwork.org/project/id/10357
• Funded by: German and Israeli Ministries of Economy
• Partner:
• Goal: Develop a high-quality, domain-adapted, real-time, tightly-integrated
media-data translator for financial sector videos
• Objectives and Technologies Approach:
• Financial news audio often includes data from mixed sources like read news
from professional speakers, spontaneous speech, noisy outbound
interviews, interviews across telephony lines, often various languages
• several speaker, audio environments, languages and domains
• Tightly integrated modular processing pipeline for ASR & SLT with
• specialized, streaming-enabled modules (e.g. robust online channel,
speaker change and language detection)
• offline target domain optimization
• online domain adaptation (e.g. financial text feeds)
9. 9 © EML European Media Laboratory GmbH Jimmy Kunzmann
EU Marie-Curie Project: LISTEN
• LISTEN: Hands-free voice-enabled interface to web applications for smart home
environments
• Project duration: 2015-2019
• Home page: http://www.listen-project.eu/
• Partner:
• Objectives and Technologies:
• Wireless acoustic sensor network (WASN) designed for smart homes
• Voice access to Internet applications (web search, dictation, social networks)
• Voice control web-enabled “smart” appliances (heating/cooling, lighting, media)
• Bridge the gap between the acoustic front-end and automatic speech recognition
research communities to enable hands-free voice interface.
• Source separation with digital microphone array and speech recognition
• First demo: http://www.listen-project.eu/demonstrations.html
10. 10 © EML European Media Laboratory GmbH Jimmy Kunzmann
EU Project: TC-STAR
• TC-STAR: Technology & Corpora for Speech to Speech Translation (SLT)
• Project duration: 2004-2007
• Goal: advance research in Automatic Speech Recognition (ASR), Spoken
Language Translation (SLT), Text to Speech (TTS) (speech synthesis)
• Objectives:
• Translation of speeches from European Parliament Plenary Sessions
• Speech recognition to perform reliably under varying speaking styles,
recording conditions, for different user communities
• Effective integration of ASR and SLT in a unique statistically sound
framework
• Expressive speech synthesis read and talk in multiple languages
• Evaluations
• Annual competitive evaluations: component technologies, full systems
• Create a technological infrastructure to foster effective delivery and to
assess scientific results